Failed to open TensorBoard for embedding visualization from Google Cloud DataLab with vector data only

Question

I plan to make an embedding visualization in TensorBoard Projector with prepared vector data (not that trained by TensorFlow), from a notebook in Google Cloud DataLab (instead of uploading TSV files in a web browser).

I've tried the code provided in this tutorial:

from tensorflow.contrib.tensorboard.plugins import projector

LOG_DIR='test_log'

# Create randomly initialized embedding weights which will be trained.
N = 10000 # Number of items (vocab size).
D = 200 # Dimensionality of the embedding.
embedding_var = tf.Variable(tf.random_normal([N,D]), name='word_embedding')

# Format: tensorflow/tensorboard/plugins/projector/projector_config.proto
config = projector.ProjectorConfig()

# You can add multiple embeddings. Here we add only one.
embedding = config.embeddings.add()
embedding.tensor_name = embedding_var.name
# Link this tensor to its metadata file (e.g. labels).
# embedding.metadata_path = os.path.join(LOG_DIR, 'metadata.tsv')

# Use the same LOG_DIR where you stored your checkpoint.
summary_writer = tf.summary.FileWriter(LOG_DIR)

# The next line writes a projector_config.pbtxt in the LOG_DIR. TensorBoard will
# read this file during startup.
projector.visualize_embeddings(summary_writer, config)

While LOG_DIR is an empty folder in the same folder with the notebook file.

Since metadata is not required in embedding visualization, I didn't set embedding.metadata_path .

Then I run the code:

from google.datalab.ml import TensorBoard as tb
tb.start('test_log')

A new page of TensorBoard can be opened, but it says:

No checkpoint was found.

when I switch to Projector view.

But as the code above shows, the data is created randomly, there should not be any checkpoint file.

Furthermore, at the next stage, I need to make an embedding visualization with my own vector data which is not trained by Tensorflow, without any checkpoint file.

When using Projector in a web browser, only a TSV file of vector data is required, it doesn't require any checkpoint file.

So the question is: What is the correct way to make embedding visualization in TensorBoard Projector from Google Cloud DataLab, with only a dataset of vector?

Thanks.

Answer 1

Looking at that tutorial, you need to run the code to periodically save off checkpoints. Those should be the basis for the information showing up in your tensor board.

saver = tf.train.Saver()
saver.save(session, os.path.join(LOG_DIR, "model.ckpt"), step)